2installations

v0.1.14

llm-html

CMDOP Skill — installation et utilisation via CMDOP agent:
Text
cmdop-skill install llm-html
Loading code...

Nettoyage HTML optimisé pour LLM : extraction d’hydratation, budgets de jetons, formats de sortie multiples.

Installation

Bash

pip install llm-html

Loading code...

Démarrage rapide

Python

from llm_html import HTMLCleaner, CleanerConfig, OutputFormat

# Nettoyage de base

 cleaner = HTMLCleaner()
result = cleaner.clean(html)
print(f"Réduction : {result.stats.reduction_percent}%")

# Hydratation première (extrait les données SSR de Next.js, Nuxt, etc.)

if result.hydration_data:
    data = result.hydration_data
else:
    cleaned = result.html

Loading code...

Fonctions de commodité

Python

from llm_html import clean, clean_to_json, clean_html, clean_for_llm

# Nettoyage rapide

result = clean(html)

# Obtenir JSON si données SSR disponibles, sinon HTML nettoyé

 data = clean_to_json(html)

# Pipeline avec contrôle complet

result = clean_html(html, max_tokens=5000)
result = clean_for_llm(html, output_format="markdown")

Loading code...

Formats de sortie

Python

from llm_html import to_markdown, to_aom_yaml, to_xtree

md = to_markdown(html)
aom = to_aom_yaml(html)
xtree = to_xtree(html)

Loading code...

Downsampling

Ciblage de budget de jetons avec l’algorithme D2Snap :

Python

from llm_html import downsample_html, estimate_tokens

tokens = estimate_tokens(html)
if tokens > 10000:
    html = downsample_html(html, target_tokens=8000)

Loading code...

Chunking sémantique

Diviser les pages importantes en morceaux LLM :

Python

from llm_html import SemanticChunker, ChunkConfig

config = ChunkConfig(max_tokens=8000, max_items=20)
chunker = SemanticChunker(config)
result = chunker.chunk(soup)
for chunk in result.chunks:
    process(chunk.html)

Loading code...

Shadow DOM

Flatter les composants Web pour la visibilité LLM :

Python

from llm_html import flatten_shadow_dom

flat = flatten_shadow_dom(html)

Loading code...

Aide

Python

from llm_html import html_to_text, extract_links, extract_images, json_to_toon

text = html_to_text(html)
links = extract_links(html, base_url="https://example.com")
images = extract_images(html)
toon = json_to_toon({"key": "value"})

Loading code...

Licence

MIT

Text

cmdop-skill install llm-html

Bash

pip install llm-html

Python

from llm_html import HTMLCleaner, CleanerConfig, OutputFormat

# Nettoyage de base

 cleaner = HTMLCleaner()
result = cleaner.clean(html)
print(f"Réduction : {result.stats.reduction_percent}%")

# Hydratation première (extrait les données SSR de Next.js, Nuxt, etc.)

if result.hydration_data:
    data = result.hydration_data
else:
    cleaned = result.html

Python

from llm_html import clean, clean_to_json, clean_html, clean_for_llm

# Nettoyage rapide

result = clean(html)

# Obtenir JSON si données SSR disponibles, sinon HTML nettoyé

 data = clean_to_json(html)

# Pipeline avec contrôle complet

result = clean_html(html, max_tokens=5000)
result = clean_for_llm(html, output_format="markdown")

Python

from llm_html import to_markdown, to_aom_yaml, to_xtree

md = to_markdown(html)
aom = to_aom_yaml(html)
xtree = to_xtree(html)

Python

from llm_html import downsample_html, estimate_tokens

tokens = estimate_tokens(html)
if tokens > 10000:
    html = downsample_html(html, target_tokens=8000)

Python

from llm_html import SemanticChunker, ChunkConfig

config = ChunkConfig(max_tokens=8000, max_items=20)
chunker = SemanticChunker(config)
result = chunker.chunk(soup)
for chunk in result.chunks:
    process(chunk.html)

Python

from llm_html import flatten_shadow_dom

flat = flatten_shadow_dom(html)

Python

from llm_html import html_to_text, extract_links, extract_images, json_to_toon

text = html_to_text(html)
links = extract_links(html, base_url="https://example.com")
images = extract_images(html)
toon = json_to_toon({"key": "value"})

llm-html

llm-html

Installation

Démarrage rapide

Fonctions de commodité

Formats de sortie

Downsampling

Chunking sémantique

Shadow DOM

Aide

Licence

Avis